Unexploded Ordnance Recognition Based On BP Neural Network

At present,with the rapid development of science and technology, people’sliving standard is continuously changing to a higher level. We should also notice thatthere are also millions of people suffering from the pains brought by war in someplace of our world. As is known to all, the world peace and development has becomethe theme of modern times, so we should also pay constantly attention to ensure thesafety of people. So far, it still has plenty of unexploded ordnance left because of themilitary war, threatening the safety of human life and property. Therefore it is ofgreat significance to detect and identify the residual unexploded ordnance.There are many conventional methods to detect and identify the unexplodedordnance. The first task of identifying the unexploed ordnance is to detectunexploded ordnance, and then to identify or exclude them. Although sometimespeople can identify the unexploded ordnance by the observation of shape or textnumber, the unexploded ordnance that can be observed in people’s visual range islittle or nothing. Common unexploded ordnance is generally not exposed to the placethat is easy to be seen, that is to say, most of them are hidden under the ground, butgenerally they will be buried in the location of the shallow ground. On the onehand,it offers the possibility to detect the location of unexploded ordnance,on theother hand it also increases the danger produced by unexploded ordnance when it isblasting.Therefore,it is quite significant to study of the detection technology andidentification method of unexploded ordnance.In the process of researching on the recognition method of unexplodedordnance, firstly we made a carefully analysis of foreign and domestic technology and methods applied on detecting unexploded ordnance, then we combined with therecogniton recognition system of the unexploded ordnance and analyzed theadvantages and disadvantages of other common methods, finally we designed theidentification system that was meeting the requirements based on above it.Accordingto the requirements of the project, we designed a new method that is more in linewith the expected objectives based on BP neural network,which is to identify theunexploded ordnance. The paper introduced the basic principle of BP neural networkin detail, and made a theoretical analysis of the important parameters in the modeland the influence of these parameters on the curve recognition. Based on the aboveresearch, the effectiveness and feasibility of this method in the detection ofunexploded ordnance is proved by repeated experiments.To realize this method, the paper analyzed some important parameters of the BPneural network, and then evaluated the four kinds of improved BP algorithmincluding OSS algorithm, RP algorithm, SCG algorithm and GDX algorithm whichare commonly used. Based on the simulation experiments of the standard curve, wechose the more suitable method to achieve recognition curve by the RP learningalgorithm which is based on BP neural network according to the principles ofexcellent choice.Curve recognition implementation is divided into two parts: the standard curverecognition and adding noisy curve recognition. Firstly the paper used the BP neuralnetwork to establish a recognition model, and then carried out recognition simulationexperiments on four kinds of standard curve (sine wave, square wave, sawtoothwave and trapezoidal wave)after all the parameters are determined. In the simulationexperiment, the paper analyzed how the main parameters of the neural network (thenumber of hidden layer nodes and the learning algorithm) effect on the recognitioncurve. On this basis of it,we analyzed the distribution of different types of noisetheoretically, added different intensity Gauss noise to the normal curve, and then analyzed the influence of noise signal on the curve recognition by simulationexperiments. Finally,through the relevant testing experiment, this paper verified theeffectiveness and validity of the algorithm.In this paper, we took the electromagnetic detection system invented by Jilinuniversity as technical support. The system recorded electromagnetic response of thetarget under different frequencies, and to obtain the in-phase component I andquadrature component Q curve of the underground target by the collected data. Then,based on the method of BP neural network, the we tested the characteristic curve ofthe unexploded ordnance for identifying the recogniton ability. At the same time,weshowed and analyzed the measured results of unexploded ordnance under differentconditions. At last, according to the main research work and the actual researchprogress, we put forward the next step outlook of the work.